MBP 1,1-10,2 Finding a workaround for unibody MBPs w/ a bad RAM slot

[B S Magnet]

This is homologous to the recent thread on PowerBook G4s with known bad RAM slots (and getting 2GB SO-DIMMs to do the work of two 1GB sticks albeit in one slot):

I have an early-2011 A1278 MBP 13/i5/2.3GHz (Sandy Bridge), which is my Snow Leopard box. It runs with one 8GB PC3-10600 DDR3 1333MHz SO-DIMM.

Since 2013, one of the two RAM slots has been bad, leaving me capped at 8GB.

Given the logic involved in the workaround described above for PBs with one bad RAM slot, I looked at whether there were ever 16GB PC3-10600 DDR3 1333MHz SO-DIMM sticks. If there were, then I can’t find any indication of it online.

I did, however, find there was/is 16GB PC3-12800 DDR3 1600MHz sticks rated for a slightly higher clock speed, leaving me to wonder whether anyone has tried successfully:

1) to use a faster-rated speed DDR3 stick in a slower system (similar to using a PC-133 on a PC-100 iBook clamshell); AND
2) tried using a 16GB stick in any of the Core i5/i7 unibody MacBook Pros to successfully have all 16GB correctly register on a single slot (like the PowerBook G4 workaround above).

If either or both of these workarounds have been tried by one of you, I’d like to hear how it went. It would be nice to run 16GB on my otherwise-venerable MBP which I’ve used since new.

 

[B S Magnet]

This is homologous to the recent thread on PowerBook G4s with known bad RAM slots (and getting 2GB SO-DIMMs to do the work of two 1GB sticks albeit in one slot):

I have an early-2011 A1278 MBP 13/i5/2.3GHz (Sandy Bridge), which is my Snow Leopard box. It runs with one 8GB PC3-10600 DDR3 1333MHz SO-DIMM.

Since 2013, one of the two RAM slots has been bad, leaving me capped at 8GB.

Given the logic involved in the workaround described above for PBs with one bad RAM slot, I looked at whether there were ever 16GB PC3-10600 DDR3 1333MHz SO-DIMM sticks. If there were, then I can’t find any indication of it online.

I did, however, find there was/is 16GB PC3-12800 DDR3 1600MHz sticks rated for a slightly higher clock speed, leaving me to wonder whether anyone has tried successfully:

1) to use a faster-rated speed DDR3 stick in a slower system (similar to using a PC-133 on a PC-100 iBook clamshell); AND
2) tried using a 16GB stick in any of the Core i5/i7 unibody MacBook Pros to successfully have all 16GB correctly register on a single slot (like the PowerBook G4 workaround above).

If either or both of these workarounds have been tried by one of you, I’d like to hear how it went. It would be nice to run 16GB on my otherwise-venerable MBP which I’ve used since new.

UPDATE 2019.12.24:

OK, so I tentatively verified how a faster-rated stock of RAM from the same group (e.g., PC3, PC2, etc.) will/should run on a slower logic board and at the clock speed rated for that board. This is a general piece of info which doesn’t, of course, break down to individual vendors (i.e., some logic boards are a lot pickier than others, and as experience goes, we’ve definitely seen finicky behaviour from certain Macs).

This leaves me with one outstanding question — something which resident hardware tinkerers like @LightBulbFun or @dosdude1 might know:

Is the 16GB cap of a unibody i5/i7 a ceiling of capacity as defined by what the memory bus can handle in total, or is it a function of how much each “lane” (RAM slot) can handle (i.e., up to 8GB-per-slot x 2)?

This is where, shy of spending over CAD$100 on a stick of PC3-12800S 16GB RAM (as an expensive experiment/gambit I can’t really swing) to try on one [working] lane/slot, I can’t really be sure. If 16GB is the memory bus cap, then running 16GB through a single lane might be slower than 2x8GB on two lanes, but at least it would enable 16GB on my MBP.

Thanks in advance, everyone.

 

[MagicBoy]

With Sandy Bridge the memory controller is on the CPU die. Only supports 8GB per slot.

They will run 32GB, but only in 4-slot boards, like the ThinkPad W520.

 

[B S Magnet]

With Sandy Bridge the memory controller is on the CPU die. Only supports 8GB per slot.

They will run 32GB, but only in 4-slot boards, like the ThinkPad W520.

This is exactly what I needed to know. Thank you.

 

[B S Magnet]

FINAL(?) UPDATE!

What I was hoping to get out of this old thread was a thorough technical explanation, rather than solely an up/down, yes/no answer. Although the yes-or-no was adequate at the time I asked, I’m also one of those nerds who hopes to really get a better understanding behind why something does, doesn’t, might, or might not work.

This week, I revisited this old thread because I’d largely forgotten I’d posted it the first time (my brain is like a sieve). And also because there’s been a kind of parallel discussion over on the iMac forum (on a thread which will, probably within the next year or so, have a home on the Early Intel Macs forum).

This time, I finally found a thorough, technical explanation behind why Sandy Bridge, most Ivy Bridge (Ivy Bridge-E excepted), and possibly Haswell* CPUs get cranky with RAM modules larger than 8GB. I found said technical answer over on superuser, and it does a pretty good job with describing why there are hardware limits involved around a hardware design bug which was later addressed by Intel — pun not intended.

Still, there seems to be some lingering ambiguity surrounding the circumstances with Haswell CPUs, such as the 2013 iMacs and MacBook Pros, as several other references I’m not linking to here indicate that Intel rectified the bug with the Ivy Bridge-E series of chips. (Ivy Bridge-E went on sale around the same time as the first Haswell CPUs.)

It’s not clear whether the fix for Ivy Bridge-E CPUs were also applied to any Haswells. Making things even less clear, Intel’s own specs show Haswell CPUs have two memory channels supporting up to 32GB (ostensibly 16GBx2); Ivy Bridge-E have four memory channels supporting up to 64GB (ostensibly 16GBx4); and even some Sandy Bridge CPUs with two memory channels showing support for 32GB (ostensibly 16GBx2).

So I’m clearer about some of this but now left less certain about other aspects. Nevertheless, I’m posting this (probably final) update for anyone else who later runs across this kind of question.

 

[Amethyst1]

Making things even less clear, Intel’s own specs show Haswell CPUs have two memory channels supporting up to 32GB (ostensibly 16GBx2); Ivy Bridge-E have four memory channels supporting up to 64GB (ostensibly 16GBx4); and even some Sandy Bridge CPUs with two memory channels showing support for 32GB (ostensibly 16GBx2).

Since one memory channel can consist of two memory modules, these limits can be “hit” using 8 GB modules if enough slots are available.

The 2013 Mac Pro (Ivy Bridge-E) can use four 32 GB modules for 128 GB.

 

[B S Magnet]

Since one memory channel can consist of two memory modules, these limits can be “hit” using 8 GB modules if enough slots are available.

That I understand. It’s also not really at question here.

The 2013 Mac Pro (Ivy Bridge-E) can use four 32 GB modules for 128 GB.

That’s perplexing to me. Intel’s own specifications (what I linked to just before your reply) indicate that the Ivy Bridge-E’s cap is 64GB across four memory channels (no mention here of memory modules) — or, 16GB per channel.

So now I’m especially confused.

 

[Amethyst1]

Intel’s own specifications (what I linked to just before your reply) indicate that the Ivy Bridge-E’s cap is 64GB across four memory channels (no mention here of memory modules) — or, 16GB per channel.

The specs were published before 32 GB modules were available and probably never updated to accommodate them.

 

[B S Magnet]

The specs were published before 32 GB modules were available and probably never updated to accommodate them.

It appears the Xeon series needs to be held discretely from non-Xeon variants. For example, the quad-core E5-1650 in the base model can accommodate up to the peculiar max memory of 375GB across four memory channels, whilst the top-end, 12-core E5-2697v2 can accommodate up to 768GB across, again, four channels.

This would suggest, respectively, that each of the four channels in the base model late 2013 Mac Pro could accommodate up to 96GB (though only 375 of the 384 would be recognized), and the top-end Mac Pro would accommodate up to 192GB per channel. OF course, all of this is the CPU specs alone and doesn’t factor in other things like firmware and logic board design limitations.

So, frankly, I don’t know.

I’m seeing on the Intel specs pages how some of the quad-core, Sandy Bridge i7 CPUs can accommodate up to 32GB in two memory channels, whereas others can only accommodate 16GB; for Ivy Bridge, most, if not all are 32GB (despite the superuser discussion I posted indicating the memory channel design problem/bug/issue with pre-Ivy Bridge-E processors being confined to just 16GB total across two channels). If this is the case, then this would refute, barring firmware settings on the device hosting the CPU, much of what that superuser discussion outlined.

 

[Amethyst1]

I’m seeing on the Intel specs pages how some of the quad-core, Sandy Bridge i7 CPUs can accommodate up to 32GB in two memory channels, whereas others can only accommodate 16GB; […]

I checked the 2600S, 2600K and 2700K: they’re all listed as supporting 32GB. Which i7s did you find to be limited to 16GB?

[…] despite the superuser discussion I posted indicating the memory channel design problem/bug/issue with pre-Ivy Bridge-E processors being confined to just 16GB total across two channels).

I apologise if I’m not getting the point, but isn’t that bug just preventing individual modules larger than 8GB from working? So to get 32GB on Sandy Bridge or “normal” Ivy Bridge, you need four slots populated with 8GB modules (16GB per channel).

 

[B S Magnet]

I checked the 2600S, 2600K and 2700K: they’re all listed as supporting 32GB. Which i7s did you find to be limited to 16GB?

Two of the six I selected to compare on this chart: the i7-2635QM (i.e., the one in the 2011 Mac mini server) and the i7-2675QM show as having maximum support for 16GB; the other four on that chart show up to 32GB. There may be others, but these were the six I used above (and had bookmarked).

I apologise if I’m not getting the point, but isn’t that bug just preventing individual modules larger than 8GB from working? So to get 32GB on Sandy Bridge or “normal” Ivy Bridge, you need four slots populated with 8GB modules (16GB per channel).

I think you are getting the point probably better than I am — namely, that the bug in question won’t recognize chips on a module being larger than 4 gigabits. So for a single, 8GB SO-DIMM, there are 16 modules, each at 4 gigabit, for a total of 64 gigabits — or, a total of 8GiB/GB.

So if I understand this correctly, a theoretical MacBook Pro or Mac mini, with one of the Sandy Bridge or Ivy Bridge (not -E) processors, equipped with four RAM slots (2 per channel), could handle 4x8GB, totalling 32GB. Of course, no MBP or Mm is equipped like that.

The part where I get confused is when we get into the non-Xeon Ivy Bridge-E CPUs and the Haswell CPUs. Intel fixed that bug for the Ivy Bridge-E series, according to the superuser discussion (and a couple of other articles from around the time those CPUs were released in 2013). The three Core (non-Xeon) Ivy Bridge-E models Intel sold are designed with four memory channels (as opposed to two) and a capacity of 64GB.

Wait.

OK, I think I’m slowly beginning to get it. :clears throat:

Without it being specified explicitly in company specifications (note: it isn‘t) or in press releases/articles at the time the chip series was released, one simply couldn’t know Ivy Bridge-E Core CPUs can handle memory modules being as large as 8 gigabits (yielding up to 16GB [SO-]DIMMs, or 8Gbx16 modules == 16GB). But the Haswell Core CPUs probably can’t (and why the person in the other thread ran into problems running either 24 or 32GB on each memory channel on their late 2013 27-inch iMac — i.e., the Haswell model with four RAM slots, or two slots per channel).

That is: the Haswell series might not have the hardware bug fixed which permits the CPU to register and handle 8-gigabit modules on a DIMM (and why the “fix” in question indicates the reason Broadwell and later Core CPUs can handle denser RAM). But again, I’m not entirely sure here.

I feel like the brilliant mind who might be able to resolve this question definitively is @dosdude1 .

 

[Amethyst1]

Two of the six I selected to compare on this chart: the i7-2635QM (i.e., the one in the 2011 Mac mini server) and the i7-2675QM show as having maximum support for 16GB; the other four on that chart show up to 32GB. There may be others, but these were the six I used above (and had bookmarked).

I've now gone through all mobile SNB i7s. Their RAM limits are given as follows:

8 GB	16 GB	32 GB
2629M	2620M	2720QM
2649M	2640M	2760QM
2617M	2655LE	2820QM
2637M	2610UE	2860QM
2657M	2630QM	2920XM
2677M	2635QM	2960XM
	2670QM
	2675QM
	2710QE
	2715QE

So if I understand this correctly, a theoretical MacBook Pro or Mac mini, with one of the Sandy Bridge or Ivy Bridge (not -E) processors, equipped with four RAM slots (2 per channel), could handle 4x8GB, totalling 32GB. Of course, no MBP or Mm is equipped like that.

The 2011/2012 27" iMacs have four slots and can handle 32 GB. They use desktop CPUs though.

 

[B S Magnet]

I've now gone through all mobile SNB i7s. Their RAM limits are given as follows:

8 GB	16 GB	32 GB
2629M	2620M	2720QM
2649M	2640M	2760QM
2617M	2655LE	2820QM
2637M	2610UE	2860QM
2657M	2630QM	2920XM
2677M	2635QM	2960XM
	2670QM
	2675QM
	2710QE
	2715QE

The 2011/2012 27" iMacs have four slots and can handle 32 GB. They use desktop CPUs though.

Yah, that looks like what I saw on there.

Now, of course, comes the more ambiguous question around Haswell Core-series CPUs — as to whether they are capable of handling 8-gigabit modules on an SO-DIMM or if they’re similarly confined to 4-gigabit modules like its generational predecessors.

 

[Amethyst1]

Yah, that looks like what I saw on there.

That means the i7 in the 2011 MBA can only handle 8 GB... wait. That MBA was only ever offered with 4 GB.

 

[B S Magnet]

That means the i7 in the 2011 MBA can only handle 8 GB... wait. That MBA was only ever offered with 4 GB.

Only 4GB? Balderdash!

This sounds like a job for… @dosdude1 !

 

[Amethyst1]

Only 4GB? Balderdash!

It gets better (or worse?). The 2012 MBA was finally offered with 8 GB but its CPUs are listed to do... 32 GB.

Now, of course, comes the more ambiguous question around Haswell Core-series CPUs — as to whether they are capable of handling 8-gigabit modules on an SO-DIMM or if they’re similarly confined to 4-gigabit modules like its generational predecessors.

I have a Broadwell PC I could test 16GB DIMMs with. I'd have to purchase a Haswell CPU to test them with though.

 

[B S Magnet]

It gets better (or worse?). The 2012 MBA was finally offered with 8 GB but its CPUs are listed to do... 32 GB.

I wonder whether the 2012 MBA was designed with only one memory channel in use, and that channel is what got the 8GB maximum as an option…

I have a Broadwell PC I could test 16GB DIMMs with. I'd have to purchase a Haswell CPU to test them with though.

I mean, unless you were already planning to procure a Haswell-based Mac for other uses, then it’s probably not worth it.

 

[Amethyst1]

I mean, unless you were already planning to procure a Haswell-based Mac for other uses, then it’s probably not worth it.

I meant to say I have a PC that uses a Broadwell CPU (they’ve finally built the kind of oddball CPU I like LOL), so theoretically, two regular-size 16GB DDR3 modules should work to bring it to 32GB. Since Haswell CPUs use the same socket, I could get one and see if it can deal with 16GB modules too or not.

 

[dosdude1]

Only 4GB? Balderdash!

This sounds like a job for… @dosdude1 !

Unfortunately you cannot upgrade past 4GB on any 2011 or 2010 MacBook Air, despite the chipsets being capable of it. This is because Apple did not route memory address line 15 to any of the chips on the board, and left it a no connect on the MCP89 and Sandy Bridge chips. Address line 15 is required for 4Gbit chips, which are needed to achieve 8GB with only 16 chips. Now the MCP79 in the 2009 MacBook Air doesn’t support memory address line 15, so that one is chipset limited to 4GB (as it too only has 16 chips).

Now with that said, a determined enough individual COULD overcome this, by way of installing two bodge wires. The first wire would need to be connected to the memory address line 15 pin of the CPU (Sandy Bridge) or MCP89 chipset (yes this would require lifting the BGA and running said wire under it to the correct pad), and connected to the address line 15 pad of EACH of the 8 memory chips on that channel. Then you’d have to repeat this with a second wire for the second channel. These wires would also need to be terminated with a pull-up resistor, as the rest of the data and address lines are. I may undertake this at some point just for fun, but it is absolutely not worth it.

 

[dosdude1]

I wonder whether the 2012 MBA was designed with only one memory channel in use, and that channel is what got the 8GB maximum as an option…



I mean, unless you were already planning to procure a Haswell-based Mac for other uses, then it’s probably not worth it.

Once again, here we run into the same issue. The MacBook Air boards only have 16 RAM chips, meaning one rank per channel (one “rank” is a combination of chips that creates a 64-bit data bus. In the case of most DDR3 machines, they use 8-bit chips, meaning 8 chips creates a single rank. The number of ranks supported is determined by the number of chip enable lines a given platform or CPU has per memory channel). I have upgraded a 2012 MBA to 16GB before, but it requires use of special “TwinDie” 8Gbit chips. These TwinDie chips are technically two chips in one, and have 4 extra pins needed for it to work properly. Luckily Apple DID route those pins (along with address line 15, which of course would be needed for their native 8GB option), so the TwinDie chips work with it. They effectively add a second rank, making it basically the same as having 32 4Gbit chips installed. The Ivy Bridge platform does NOT support “real” 8Gbit chips (which are also hard to find and very expensive).

 

[Amethyst1]

I have upgraded a 2012 MBA to 16GB before, but it requires use of special “TwinDie” 8Gbit chips.

What about the 2013/2014 Haswell MBA?

 

[dosdude1]

What about the 2013/2014 Haswell MBA?

Those are different, and use 32-bit LPDDR3. As such, only two chips make up a rank. LPDDR3 chips in that footprint were made in up to 32Gbit density, four of which make up 16GB. You can indeed install these 32Gbit chips on a 2013 or 2014 MacBook Air, and it works exactly as intended after modifying the SPD data in the system’s SPI ROM with the appropriate data. I have performed this upgrade a few times as well.

 

[Needleroozer]

Unfortunately you cannot upgrade past 4GB on any 2011 or 2010 MacBook Air, despite the chipsets being capable of it. This is because Apple did not route memory address line 15 to any of the chips on the board, and left it a no connect on the MCP89 and Sandy Bridge chips. Address line 15 is required for 4Gbit chips, which are needed to achieve 8GB with only 16 chips. Now the MCP79 in the 2009 MacBook Air doesn’t support memory address line 15, so that one is chipset limited to 4GB (as it too only has 16 chips).

Now with that said, a determined enough individual COULD overcome this, by way of installing two bodge wires. The first wire would need to be connected to the memory address line 15 pin of the CPU (Sandy Bridge) or MCP89 chipset (yes this would require lifting the BGA and running said wire under it to the correct pad), and connected to the address line 15 pad of EACH of the 8 memory chips on that channel. Then you’d have to repeat this with a second wire for the second channel. These wires would also need to be terminated with a pull-up resistor, as the rest of the data and address lines are. I may undertake this at some point just for fun, but it is absolutely not worth it.

Just to confirm - I have a 4GB late-2013 A1502 13” retina MacBook Pro, and my read of the schematics and boardview (820-3536) indicates that the same situation applies - address line 15 is N/C for both banks even though one could buy dual-die 8Gb DDR3L chips that would fit in the FBGA-96 footprint.

Is that correct? After seeing your RAM upgrade videos on YouTube I was hopeful that I could upgrade this laptop’s RAM, but I’m not sure I’m up for reballing the CPU, especially since I’d only get 8GB if it all worked out.

 

[dosdude1]

Just to confirm - I have a 4GB late-2013 A1502 13” retina MacBook Pro, and my read of the schematics and boardview (820-3536) indicates that the same situation applies - address line 15 is N/C for both banks even though one could buy dual-die 8Gb DDR3L chips that would fit in the FBGA-96 footprint.

Is that correct? After seeing your RAM upgrade videos on YouTube I was hopeful that I could upgrade this laptop’s RAM, but I’m not sure I’m up for reballing the CPU, especially since I’d only get 8GB if it all worked out.

Luckily that's not an issue with that model, as it uses 16-bit chips instead of the standard 8-bit chips, so address line 15 isn't needed. However, what is an issue is that the 4 extra pins needed for TwinDie chips are NOT routed... Meaning your only chance would be to use actual 8Gbit chips and hope they work. These are the chips you'd need. I really have no idea if Haswell U-series supports 8Gbit chips or not, but I may test with my machine at some point if you don't end up doing so beforehand. If you do decide to attempt this upgrade, let me know, and I can create the necessary SPD data for you.

 

[Needleroozer]

Luckily that's not an issue with that model, as it uses 16-bit chips instead of the standard 8-bit chips, so address line 15 isn't needed. However, what is an issue is that the 4 extra pins needed for TwinDie chips are NOT routed... Meaning your only chance would be to use actual 8Gbit chips and hope they work. These are the chips you'd need. I really have no idea if Haswell U-series supports 8Gbit chips or not, but I may test with my machine at some point if you don't end up doing so beforehand. If you do decide to attempt this upgrade, let me know, and I can create the necessary SPD data for you.

Thanks for the offer!
I found a set of these chips for pretty cheap on Aliexpress and will probably order them to try it out - I will definitely post here with progress if I do.
(I first scrolled through twenty pages of eBay SODIMMs trying to find 8GB sticks with eight FBGA-96 chips, but no luck there...)

Out of curiosity, what four unrouted pins are necessary for TwinDie chips? The 820-3536 schematic’s NC pins seem to match the NC pins on the TwinDie datasheet I pulled up for an MT41K512M16VRP, but I’m sure I’m just missing something again.